Identifying cancer cell-state transitions from multimodal single-cell data
Phenotypic plasticity allows cancer cells to evade therapy, yet the transient nature of state transitions has made their molecular drivers difficult to define. Here, we present a single-cell framework that leverages the temporal delays between mRNA and protein accumulation to directly capture cells undergoing phenotypic switching. We show that differentiation-associated delays between transcript and protein accumulation are detectable in multimodal single-cell data as discordant mRNA and surface-protein levels. Applying this strategy to the K562 leukemia model, which alternates between differentiated CD24- and progenitor-like CD24+ states, we identify transitioning cells and derive a transcriptional signature linking cell-cycle progression and mitochondrial remodeling to plasticity. Genome-wide CRISPR screening confirms key regulators of plasticity, including BCR-ABL1 and mitochondrial homeostasis genes. We summarize the transition-associated program into a score that predicts imatinib response in chronic myeloid leukemia, stratifies survival in acute myeloid leukemia, and retains prognostic value across 31 TCGA tumor types. Spatial transcriptomics reveals localized plasticity hotspots in solid tumors. Together, this framework exposes the molecular basis of cancer plasticity and enables its quantification across tumors.